Large-Scale Optical Neural Networks based on Photoelectric Multiplication

By Ryan Hamerly, Alexander Sludds, Liane Bernstein, Marin Soljačić, Dirk Englund. Physical Review X, 2018

413 citations recorded by the source corpus.

Recent success in deep neural networks has generated strong interest in hardware accelerators to improve speed and energy consumption. This paper presents a new type of photonic accelerator based on coherent detection that is scalable to large ($N \gtrsim 10^6$) networks and can be operated at high (GHz) speeds and very low (sub-aJ) energies per multiply-and-accumulate (MAC), using the massive spatial multiplexing enabled by standard free-space optical components. In contrast to previous approaches, both weights and inputs are optically encoded so that the network can be reprogrammed and trained on the fly. Simulations of the network using models for digit- and image-classification reveal a "standard quantum limit" for optical neural networks, set by photodetector shot noise. This bound, which can be as low as 50 zJ/MAC, suggests performance below the thermodynamic (Landauer) limit for digital irreversible computation is theoretically possible in this device. The proposed accelerator can implement both fully-connected and convolutional networks. We also present a scheme for back-propagation and training that can be performed in the same hardware. This architecture will enable a new class of ultra-low-energy processors for deep learning.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/a5feaf3d02d4312bb9d309d7746e46c87aa63176

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

Citations in the corpus

Parallel convolutional processing using an integrated photonic tensor core (2021, background): “…Fourier transforms, convolu-tions, matrix multiplication and so on) has existed for many decades 26 and continues to inspire computing architectures 27–30 , precise control of the optical phase over the entire system remains the primary factor limiting scalability and commercialization of free…”
Experimental photonic quantum memristor (2021, background): “[18] Ryan Hamerly, Liane Bernstein, Alexander Sludds, Marin Soljačić, and Dirk Englund, “Large-scale optical neural networks based on photoelectric multiplication,” Physical Review X 9, 021032 (2019).”
Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data (2019, background): “…DNN will be signiﬁcantly promoted along with the rapid advances in improvement of DNN training eﬃciency, e.g., a recent study has suggested that a novel photonic chip has a potential to be used to train deep neural networks 10-million of times more eﬃciently than current CPUs/GPUs do [77].”
An optical neural chip for implementing complex-valued neural network (2021, background): “A highly parallelized optical neural network accelerator based on photo-electric multiplication has been reported 31,32 , which is also designed for real arithmetic because the optical signals had already been converted to photocurrents before reaching the accumulator.”
Programmable phase-change metasurfaces on waveguides for multimode photonic convolutional neural network (2020, methodology): “Their use in programmable photonic devices thus can have a signi ﬁ cant advantage in power consumption over electro-optic 27 – 29 or thermo-optic methods 30 – 32 .”
Space-efficient optical computing with an integrated chip diffractive neural network (2022, neutral): “However, these free-space approaches require heavy auxiliary or modulation equipment which is space-consuming 34,35 and exceedingly challenging to program in real-time 31,36,37 , hindering their feasibilities towards the large-scale implementation and commercialization of photonic neural networks.”